人工智能培训

搜索

机器学习论文:类别级6D对象姿态和大小估计的归一化对象坐标空间(Normalized Object Coordinate Space for Category-

[复制链接]
fanjz 发表于 2019-1-11 11:03:53 | 显示全部楼层 |阅读模式
fanjz 2019-1-11 11:03:53 104 0 显示全部楼层
机器学习论文:类别级6D对象姿态和大小估计的归一化对象坐标空间(Normalized Object Coordinate Space for Category-Level 6D Object Pose and  Size Estimation)本文的目的是估计RGB-D图像中的非显示对象实例的6D姿势和尺寸。与“实例级”6D姿势估计任务相反,我们的问题假设在训练或测试时间内没有确切的对象CAD模型可用。为了处理给定类别中的不同和看不见的对象实例,我们引入了规范化对象坐标空间(NOCS) -  - 为类别中的所有可能对象实例共享规范表示。然后训练基于Ourregion的神经网络直接推断从观察像素到该共享对象表示(NOCS)的对应性以及诸如类标签和实例掩模之类的其他对象信息。这些预测可以与深度图组合以联合估计主题6D在杂乱的场景中的多个对象的姿势和尺寸。为了统一我们的网络,我们提出了一种新的上下文感知技术,以生成大量完全注释的混合现实数据。为了进一步改进我们的模型并评估其在实际数据上的性能,我们还提供了具有大环境和实例变化的完全注释的真实世界数据集。大量实验表明,所提出的方法能够在重新环境中进行环节估计看不见的对象实例的姿态和大小。还在标准的6Dpose估算基准上实现了最先进的性能。
The goal of this paper is to estimate the 6D pose and dimensions of unseenobject instances in an RGB-D image.Contrary to "instance-level" 6D poseestimation tasks, our problem assumes that no exact object CAD models areavailable during either training or testing time.To handle different and unseen object instances in a given category, weintroduce a Normalized Object Coordinate Space (NOCS)---a shared canonicalrepresentation for all possible object instances within a category.Ourregion-based neural network is then trained to directly infer thecorrespondence from observed pixels to this shared object representation (NOCS)along with other object information such as class label and instance mask.These predictions can be combined with the depth map to jointly estimate themetric 6Dpose and dimensions of multiple objects in a cluttered scene.Totrain our network, we present a new context-aware technique to generate largeamounts of fully annotated mixed reality data.To further improve our model andevaluate its performance on real data, we also provide a fully annotatedreal-world dataset with large environment and instance variation.Extensive experiments demonstrate that the proposed method is able torobustly estimate the pose and size of unseen object instances in realenvironments whilealso achieving state-of-the-art performance on standard 6Dpose estimation benchmarks.机器学习论文:类别级6D对象姿态和大小估计的归一化对象坐标空间(Normalized Object Coordinate Space for Category-Level 6D Object Pose and  Size Estimation) f6b0zZAIG6Brda0k.jpg
URL地址:https://arxiv.org/abs/1901.02970     ----pdf下载地址:https://arxiv.org/pdf/1901.02970    ----机器学习论文:类别级6D对象姿态和大小估计的归一化对象坐标空间(Normalized Object Coordinate Space for Category-Level 6D Object Pose and  Size Estimation)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

fanjz当前离线
新手上路

查看:104 | 回复:0

快速回复 返回顶部 返回列表